Fix AI provider exclusivity, provider-aware models, and Gemini streaming

[shettydev]

Summary

• add persistent preferences.activeProvider (openrouter/gemini) and return it from /ai/settings
• enforce single active provider by clearing the opposite BYOK key on save and normalizing active model/provider state
• make /ai/models provider-aware with contract { provider, mode, models } and Gemini model support
• route conversation send flow through active provider and add Gemini processing path in queue worker
• align Gemini SSE emissions with OpenRouter (processing, progress, message user+assistant, complete)
• add Redis-backed stream fanout bridge in StreamService so worker/API process separation still delivers SSE
• add web fallback polling after send success to prevent stale UI when SSE is delayed
• update settings UI + store + API typings for unambiguous active provider/model behavior

Tests

• bun run test -- src/modules/ai/__tests__/ai.controller.spec.ts src/modules/conversations/services/__tests__/queue.gemini-stream.spec.ts src/modules/conversations/services/__tests__/queue.service.spec.ts src/modules/conversations/__tests__/conversation.controller.spec.ts (in packages/mukti-api)
• bun run test -- src/lib/hooks/__tests__/use-conversations.spec.ts (in packages/mukti-web)